Respondent-Driven Sampling

Overview

RDS is a type of snowball sampling used for analyzing characteristics of hidden or hard-to-reach populations. It was developed in 1997 by Dr. Douglas Heckathorn, a professor of Sociology at Cornell and has been applied to groups ranging from men who have sex with men, injection drug users, children living on the street and jazz musicians. RDS relies on multiple waves of peer-to-peer recruitment and statistical adjustments to try and approximate random sampling. The extent to which RDS-derived estimates are valid and generalizable remains a source of controversy in the peer-reviewed literature.

Description

RDS sampling consists of the following three steps:

1. Seed selection: All RDS studies begin with a small number of seeds from the target population (e.g., 3-15 people). Seeds should be diverse and well-networked, but they do not need to be chosen randomly.

2. Interviews and recruitment: Seeds complete the interview process and receive a predetermined number of coupons that they can use to recruit other people like them (Wave 1). The recruits of Wave 1 then complete the interview process and recruit Wave 2. This referral chain continues until the desired sample size is reached.

3. Incentives: Participants receive two incentives: one for completing the interview, and one for each peer that is successfully recruited.

RDS only works in populations that are connected to one another. Furthermore, the population has to be large enough to sustain long referral chains without repeated participants.

Once the sample has been recruited, statistical techniques have been developed to try and reduce biases in the data. RDS inference or analysis focuses on two main sources of bias:

1. Differential social network sizes: People with small social networks are weighted more heavily than people with large social networks to compensate for the fact that people with small networks are likely underrepresented.

2. Differential recruitment: People whose probability of recruitment is artificially increased due to homophily (e.g., same race as the recruiter) are weighted less than people who may be left out of the sample simply because they have certain characteristics that are different than the recruiters.

Finally, calculations are derived using a form of RDS Estimator:

• RDS I Estimator:

•  

  RDS II Estimator:

where,

• Px = estimated proportion of group X

• Sxy = proportion of group X selected by group Y

• Syx = proportion of group Y selected by group X

• D = overall average network size

• Dx = average network size (degree) for group X

• Dy = average network size (degree) for group Y

• nx = number of respondents in group X

• n = total number of respondents

RDS I and RDS II estimators tend to yield similar results when data-smoothing is used. The main difference between the two different methods relates to variance, with RDS I using a bootstrap technique and RDS II using an analytical technique. The RDS I estimator has a tendency to underestimate variance, and the RDS II estimator has a tendency to overestimate variance. Improving the way that variance (and sample size) are calculated is an area of active research in RDS.

Readings

Textbooks & Chapters

Wejnert, C, Heckathorn, D. Chapter 22: Respondent-Driven Sampling: Operational Procedures, Evolution of Estimators, and Topics for Future Research. In Williams, M, Vogt, W. (Eds.), The SAGE Handbook of Innovation in Social Research Methods. p. 473-498. London: SAGE Publications, Ltd., 2011. doi: http://dx.doi.org/10.4135/9781446268261.n27
Comprehensive overview. A great place to start if you’re new to RDS.

Johnston, LG. Introduction to HIV/AIDS and Sexually Transmitted Infection Surveillance, Module 4: Introduction to Respondent Driven Sampling. WHO: Geneva, Switzerland, 2013.http://applications.emro.who.int/dsaf/EMRPUB_2013_EN_1539.pdf
Written for a lay audience. Oriented towards implementation.

Methodological Articles

Heckathorn, DD. Respondent-Driven Sampling: A New Approach to the Study of Hidden Populations. Soc Probl. 1997;44:174-199. doi: http://dx.doi.org/10.2307/3096941.
Seminal RDS article.

Goel, S, Salganik, MJ. Assessing respondent-driven sampling. PNAS. 2010;107:6743-6747. doi:http://dx.doi.org/10.1073/pnas.1000261107.
RDS simulation, with special attention to variance.

McCreesh, N, Frost, SDW, Seeley, J, et al. Evaluation of respondent-driven sampling.Epidemiology. 2012;23:138-147. doi: http://dx.doi.org/10.1097/EDE.0b013e31823ac17c. 
Empirical comparison of RDS estimates to total population data.

Rudolph, AE, Fuller, CM, Latkin, C. The Importance of Measuring and Accounting for Potential Biases in Respondent-Driven Samples. AIDS Behav. 2013;17:2244-2252. doi:http://dx.doi.org/10.1007/s10461-013-0451-y.
Discussion of assumptions and potential biases of RDS, with recommendations for future research.

Salganik MJ, Heckathorn DD. Sampling and Estimation in Hidden Populations Using Respondent-Driven Sampling. Sociol Methodol. 2004;34:193–240. doi:http://dx.doi.org/10.1111/j.0081-1750.2004.00152.x
Proof that the RDS-I estimator is asymptotically unbiased.

Salganik MJ. Commentary: Respondent-driven Sampling in the Real World. Epidemiology. 2012;23:148-150. doi: http://dx.doi.org/10.1097/EDE.0b013e31823b6979
Commentary on McCreesh et. al., with a call for more comparative studies.

Volz E, Heckathorn DD. Probability based estimation theory for respondent driven sampling. J Off Stat. 2008;24:79. https://www.scb.se/contentassets/ff271eeeca694f47ae99b942de61df83/probability-based-estimation-theory-for-respondent-driven-sampling.pdf. 
Foundations for RDS-II estimator.

Application Articles

Implementation and Analysis of Respondent Driven Sampling: Lessons Learned from the Field.J. Urban Health. 2006;83.1S. http://link.springer.com/journal/11524/83/1/suppl/page/1
Special supplement issue devoted to RDS. Applications to drug users, sex workers and men who have sex with men.

Johnston, LG, Thurman, TR, Mock, N, et al. Respondent-driven sampling: A new method for studying street children with findings from Albania. Vulnerable Children and Youth Studies. 2010;5:1-11. https://www.tandfonline.com/doi/abs/10.1080/17450120903193923?journalCode=rvch20
Application of RDS to assess characteristics of street children in Albania.

Ma, X, Zhang, Q, He, X, et al. Trends in prevalence of HIV, syphilis, hepatitis C, hepatitis B, and sexual risk behavior among men who have sex with men: results of 3 consecutive respondent-driven sampling surveys in Beijing, 2004 through 2006. JAIDS. 2007;45:581-587. doi:http://dx.doi.org/10.1097/QAI.0b013e31811eadbc.
Application of RDS to assess trends in STIs among men who have sex with men in China.

Shahmanesh, M, Wayal, S, Cowan, F, et al. Suicidal behavior among female sex workers in Goa, India: the silent epidemic. AJPH. 2009;99:1239-1246. doi:http://dx.doi.org/10.2105/AJPH.2008.149930.
Application or RDS to assess suicidal behavior in female sex workers in India.

Software

RDSAT: www.respondentdrivensampling.org
This website is maintained by Dr. Douglas Heckathorn, the father of RDS. It hosts a link for downloading RDSAT, as well as the RDSAT User Manual. The website also contains some general background on RDS and links to “core RDS references,” mainly by Dr. Heckathorn. The references don’t seem as up-to-date as those on Drs. Johnston’s and Salganik’s pages (below).

RDS Analyst: http://hpmrg.org/
This website is maintained by the Hard-to-Reach Population Methods Research Group, which includes several prominent RDS researchers. It hosts a link for downloading RDS Analyst, as well numerous tutorials and guides on installing and using the software. The website also contains instructions on how to join the RDS Analyst Discussion and Help Forum. In addition, relevant trainings and workshops may be advertised here.

NetDraw: https://sites.google.com/site/ucinetsoftware/download?authuser=0
This is where the NetDraw graphing software can be freely downloaded.

Johnston, LG. Introduction to HIV/AIDS and Sexually Transmitted Infection Surveillance, Module 4 Supplement: A Guide to Using RDS Analyst and NetDraw. WHO: Geneva, Switzerland, 2013.
http://applications.emro.who.int/dsaf/EMRPUB_2014_EN_1686.pdf
A simple step-by-step guide to analysis using RDS Analyst and NetDraw.

Websites

http://globalhealthsciences.ucsf.edu/pphg/gsi/epidemiologic-surveillance/ibbs-toolbox
UCSF has created a comprehensive Toolbox for Conducting Integrated HIV Bio-Behavioral Surveillance (IBBS) in Key Populations. The tool box contains sample protocols, budgets, field team job descriptions, operations manuals and qualitative interview guides for RDS.

www.lisagjohnston.com
Dr. Lisa Johnston is a consultant specializing in RDS and RDS training. Most documents use very accessible, non-technical language. Her PhD is in International Health and Development, so many of her applications are skewed to the public health realm.

www.princeton.edu/~mjs3/rds.shtml
Dr. Matthew Salganik’s website is more academic than Dr. Johnston’s, but not quite as current. Dr. Salganik maintains an RDS listserv, and instructions for signing up are on his page. The listserv is a good way to learn about upcoming trainings and workshops on RDS.

Courses

European Consortium for Political Research: 2014 Summer School in Methods and Techniques
http://ecpr.eu/Events/EventDetails.aspx?EventID=92
This is a one-week introductory course about RDS, to be held from July 24 – August 9, 2014 at the University of Ljubljana, Slovenia. The course will be taught by Dr. Lisa Johnston.